volume 3, issue 3, september 2013 an automatic delaunay ... 3/issue 3/ijeit1412201309_22.pdf ·...

ISSN: 2277-3754

ISO 9001:2008 Certified International Journal of Engineering and Innovative Technology (IJEIT)

Volume 3, Issue 3, September 2013

134

Abstract—This paper presents a density-controlled Delaunay

triangulation system for finite element discretizations of practical

engineering problems and scientifically interacting mechanisms.

B-splines are constructed to produce boundary points on the

curved boundaries of the geometry according to the required point

spacing. The lost boundaries in concave and thin geometric

features are recovered through recursively inserting

pseudo-points on the lost edges. The determinant criterion and

modes for removing pseudo-points are correspondingly

established as well. Two methods are employed to add interior

points: direct method and pre-test method. For the geometry with

several sub-domains, corresponding measures for handling

overlapped edges are established to ensure the mesh compatibility

and element distribution between adjacent sub-domains.

Index Terms—Boundary integrity, Delaunay, Point spacing,

Voronoi.

I. INTRODUCTION

Finite element method is a general purpose method for

numerical analysis in engineering applications and scientific

investigations. It is an important component of

computer-aided design and manufacture, and extensively

applied in the fields of hydraulics, hydrology, water resources,

geotechnical engineering, solid/fluid mechanics, computer

graphics, mechanics, and architecture [1], etc. The major

characteristic of finite element method is geometry discretion

and slice interpolation, i.e. finite element mesh generation. In

two-dimensional case, finite element mesh contains two types,

triangular mesh and quadrilateral mesh. Among the numerous

kinds of triangular mesh generation methods, Delaunay

triangulation method is the most widely used one. Delaunay

criterion was first proposed by the Russian mathematician,

Delaunay [2], in 1934. However, it was not until the early

1980s that Delaunay triangulation method was proposed by

Lawson [3] and Watson [4]. Delaunay triangulation method

has two crucial criteria: maximum-minimum angle criterion

and in-circle criterion. The maximum-minimum angle

criterion [5] refers to that the sum of minimal internal angles

of all the triangles obtains the maximal value and that of

maximal internal angles obtains the minimal value. This

criterion enables Delaunay triangulation to automatically

avoid generating long and thin elements with small internal

angles in two-dimensional case. It not only can guarantee the

convergence of the computation, but also ensure that the

meshing result is optimal. The in-circle criterion signifies that

the circumcircle of each triangle during Delaunay

triangulation does not contain any other points.

Bowyer-Watson algorithm [4], [6] just takes advantage of the

in-circle criterion. Delaunay triangulation algorithm has

excellent flexibility and extensibility. It is easily extended to

three-dimensional case, so it has attracted many research

scholars [7]–[9]. The conventional Delaunay triangulation

technique has been relatively mature. The recent research

mainly focused on the boundary integrity algorithm of

constrained Delaunay triangulation [10], [11], as well as how

to overcome the thin elements derived from the degeneration

phenomenon of Bowyer-Watson approach. Since the

Delaunay triangulation method expresses sufficient

effectiveness only for convex geometric features but cannot

ensure the boundary integrity for concave features, it is

necessary to introduce a boundary recovery procedure when

performing Delaunay triangulation to concave domains. This

paper adopted Delaunay triangulation method to generate

two-dimensional triangular element meshes and developed an

automatic triangle meshing system based on point spacing. An

intensive study was made on the insertion of boundary points

and interior points, boundary integrity, and treatment of

overlapped boundaries for multi-domain geometries. Here,

the main research content of this paper would be illustrated in

detail in the following sections.

II. THEORETICAL FOUNDATION

A. The Voronoi diagram

The implementation of Delaunay triangulation completely

relies on the Voronoi diagram. The concept of Voronoi was

discovered by the Russian mathematician, Voronoi, in 1908.

Suppose given a set of points S in a plane, including n

distinct points. For each point iP , its Voronoi territory,

iV P , is defined as the locus of points in the plane whose

distance to iP is closer than to any other points of S .

/ , , , , ,i i j i jV P P d P P d P P P P S j i (1)

The Voronoi cell of each point in S can be expressed as a

convex polygon. Thus, the whole plane where S is located is

divided into n Voronoi polygons, each of which is

associated with a unique point in the set. The n convex

polygons form the Voronoi diagram of point set S . In the

An Automatic Delaunay Triangular Mesh

Generation Method based On Point Spacing Lu Sun

1,2,￥ Gour-Tsyh Yeh

1,* Fang-Pang Lin

3,† Guoqun Zhao

2,§

1 Graduate Institute of Applied Geology, National Central University, Taiwan

2 Engineering Research Center for Mould & Die Technologies, Shandong University, China

3Data Computing Division, National Center for High-Performance Computing, Taiwan

ISSN: 2277-3754



135

Voronoi diagram, every vertex is the common intersection of

three edges [6]. The edge of Voronoi polygon is defined as the

perpendicular bisector of the line segment joining two

adjacent points. When point iP lies on the boundary of the

convex hull, its Voronoi polygon is unbounded. The straight

line duality of the Voronoi diagram forms Delaunay triangles

[2]. Each Voronoi vertex corresponds to a unique Delaunay

triangle and vice versa. For every Voronoi vertex, the

circumcircle contains no other points of set S .

B. B-splines

B-spline is short for basis spline, first proposed by

Schoenberg [12]. In this paper, B-splines are used to

interpolate boundary points for the geometries with curved

boundaries.

Assume uP represents the position vector of an

arbitrary point in a curved boundary. Then, the -thk order

B-spline along variable u is defined by

,

0

( ) ( )n

i i k

i

u N u

P B (2)

where iB indicates the position vectors of the control points,

with the number equal to the original points, 1n . , ( )i kN u

denotes the normalized B-spline blending functions. While a

-thk order B-spline is constructed, the Cox-de Boor formula

[13], [14] of the -thi blending function, , ( )i kN u , is defined

as

1

,1

1 ,( )

0 otherwise

i i

i

u u uN u

(3)

, , 1 1, 1

1 1

( ) ( ) ( )i i ki k i k i k

i k i i k i

u u u uN u N u N u

u u u u

(4)

where iu represents the knots of the B-spline, with

1i iu u .

III. THE IMPLEMENTATION PROCEDURES AND

KEY TECHNIQUES OF DELAUNAY

TRIANGULATION

Fig. 1 shows the whole implementation process of the

triangle meshing system developed in this paper. In the

following, the procedures of Voronoi diagram construction

and Delaunay triangulation are introduced concretely by

taking the geometry shown in Fig. 2(a) as an example.

Fig. 1. The procedures of Delaunay triangulation

(a) The input data of the geometry

(b) The initial voronoi diagram

Input geometry

Output final mesh

Construct a simple rectangular structure and create the initial Voronoi diagram

Generate boundary points based on specified spacing of input points

Construct the Voronoi diagram and Delaunay triangular mesh for boundary points

Boundary integrity

Generate interior points and insert them into the present Voronoi diagram

Smooth the mesh

ISSN: 2277-3754



136

(c) Identification of boundary nodes

(d) Mesh after inserting boundary nodes

Fig. 2. The generation of boundary points

A. Input data

The meshing system developed in this paper requires two

aspects of input data: data points and boundary edges. For the

geometry in Fig. 2(a), there are 12 data points, denoted as 1P ,

2P … 12P , with the number 12np . The decimal in the

parentheses after the point label represents the point spacing

specified by users. The geometry consists of 12 boundary

edges 1b , 2b … 12b , with the number 12nb . It should be

noticed that the input boundary edges for each meshed region

must follow such a rule: the outer edges are arranged in

counterclockwise order and the inner edges are arranged in

clockwise order.

B. Construction of initial Voronoi diagram and Delaunay

triangles

First of all, it needs to define a convex hull within which all

the input points lie. In this paper, a rectangle enveloping all

input points is defined, as seen in Fig. 2(b).

Table I. Data structures of the initial Voronoi diagram

Voronoi Forming points Neighboring Voronoi

1 2 3 1 2 3

V1 N1 N2 N3 none V2 none

V2 N1 N3 N4 none none V1

Then, construct the initial Voronoi diagram and Delaunay

triangular elements based on the rectangular structure. The

initial diagram is composed of two Voronoi vertices (1V and

2V ) and two Delaunay triangles. Table I lists the data

structures of the initial diagram [6], [7].

C. Generation of boundary points according to specified

point spacing

Before boundary-point creation, it needs to subdivide the

geometry boundaries into several segments. To achieve

correct subdivision, special corner points are suggested to be

identified first. A special corner point is defined as the point

where three or more boundary edges meet. This paper

established the segmentation rule as follows: (1) the straight

edge is considered as a single segment itself; (2) the

end-to-end curved edges are integrated to form a whole

segment; (3) if a special corner point is encountered, its

connected edges are regarded as belonging to different

segments. The boundary points are generated based on the

specified point spacing. The point spacing is reflected by the

lengths of the edges connected to the point. Taking the -thi

edge ib as an example, suppose that the two endpoints of ib

are aP and bP with specified point spacing labeled as

aps P and bps P , respectively.

First, compute the midpoint of ib , denoted as mP . If ib

belongs to a curved segment, B-spline should be created. mP

is identified as the point derived from interpolation at

( ) / 2m a bu u u by “(2)”, “(3)”, and “(4)”. Then,

compute the practical point spacing at mP by

' / 2m a m m bps P P P P P (5)

Second, check whether mP is required to be inserted.

Determine the lower value between aps P and bps P ,

denoted as minps . If '( )mps P is larger than

minps , accept mP . Otherwise, reject mP and turn to the

check for the next edge. If mP is accepted, update the data

information of relevant points and edges. And, a specified

point spacing value is assigned to mP by interpolation.

( ) ( )( )

a m b b a m

m

a m m b

ps P P P ps P P Pps P

P P P P

(6)

At the same time, edge ib is decomposed into two new edges.

Then, the above procedures are repeated over the two new

edges again until the practical point spacing are able to satisfy

the specified values. Fig. 2(c) provides the boundary contour

after identifying the inserted points based on specified point

spacing, containing 67 points and 67 edges. The numbers in

ISSN: 2277-3754



137

square boxes represent the serial numbers of the boundary

points.

D. Construction of Voronoi diagram and Delaunay

triangles of boundary points

Now, we insert all the boundary points into the initial

Voronoi diagram. Taking the point set in Fig. 3(a) as the

example, suppose that there are 11 distinct points in set S ,

1 2 3 11, , ... P P P P , and a new point 12P will be inserted to

the present Voronoi diagram. The insertion operation is

realized through the following steps [6]:

(a) the original Voronoi diagram

(b) Check position relationships (c) local Voronoi diagram

Of point and triangle

(d) Global Voronoi diagram

Fig. 3. Insertion of new point

(1) Identify the Delaunay triangle and the Voronoi structure

containing the new point. To save computing time and reduce

programming work, this paper uses the right-hand rule to

check the position relationships between the new point and

each Delaunay triangle. For an arbitrary Delaunay triangle,

say randomly selected, the introduced point is able to form

three new triangles with its three edges. For example, the

introduced point 12P divides the Delaunay triangle 1 8 7PP P

(corresponding to Voronoi vertex 1V , this triangle is selected

randomly on the first try) into three new smaller ones, as the

amplified view shown in Fig. 3(b). Then, the right-hand rule is

employed to check whether these three triangles have the

same rotations. The check result shows that the orientations of

triangles 12 1 8P PP and 12 7 1P P P are both counterclockwise,

whereas that of triangle 12 8 7P P P is different. So, Point 12P is

in the exterior of triangle 1 8 7PP P and is located in the outside

of the edge 8 7-P P . In order to save time, the follow-up search

process is performed from the vertices adjacent to 1V

(corresponding to triangle 1 8 7PP P ) and in the outer side of

edge 8 7-P P . Then, Voronoi vertex 8V and its duality

Delaunay triangle 7 8 11P P P are selected for check. As shown

in Fig. 3(b), new point 12P forms three new triangles with the

three edges of triangle 7 8 11P P P . According to right-hand rule,

these three new triangles have the same rotations of

counterclockwise. Therefore, the dual Delaunay triangle of

Voronoi Vertex 8V is just the triangle containing point 12P .

(2) Determine the Voronoi vertices to be deleted.

According to the fundamental property of Voronoi diagram

that the circumcircle of each vertex only contains its three

forming points but not involves any other points, the vertices

whose circumcircles contain the newly inserted point should

be deleted. As shown in Fig. 3(a), there are four Voronoi

vertices to be deleted, 1V , 8V , 9V , and 13V . Their dual

Delaunay triangles should be deleted as well.

(3) Construct local Voronoi diagram and Delaunay

triangles

When the identified vertices are deleted, an empty convex

polygon arises in the deleted region, as the polygon

1 8 9 10 11 7PP P P P P shown in Fig. 3(a). The new point 12P is in

the interior of the polygon. The edges 1 8-P P , 8 9-P P , 9 10-P P ,

10 11-P P , 11 7-P P , and 7 1-P P of the empty polygon are the

residual edges of deleted triangles in practice. Every residual

edge gives rise to a new Voronoi vertex with 12P . In this case,

six new vertices are constructed, as 1 'V , 2 'V , 3 'V , 4 'V ,

5 'V , and 6 'V shown in Fig. 3(c).

(4) Merge the local Voronoi diagram into the global

diagram

The merger of local Voronoi diagram contains two aspects.

The first is to add the newly generated vertices to the original

diagram directly. The second is to modify some of the

ISSN: 2277-3754



138

neighboring information of the Voronoi vertices adjacent to

the deleted vertices. Fig. 3(d) shows the new Voronoi diagram

after inserting point 12P .

(5) After one point is inserted, repeat above steps (1)-(4) to

insert the next one.

For the geometry in Fig. 2(a), the 67 boundary points (see

Fig. 2(c)) are inserted into the initial Voronoi diagram (see

Fig. 2(b)) by the method stated above. And, the Delaunay

triangles formed by all boundary points are achieved, as

shown in Fig. 2(d).

E. Boundary integrity

Make an intuitive comparison of the boundary contours

between the input geometry in Fig. 2(a) and the mesh after

inserting boundary points in Fig. 2(d). It is clearly seen that

the input edges 5b and 11b are missing. This is an inevitable

issue when performing Delaunay triangulation for a given set

of boundary edges which are defined by their endpoints. It

usually occurs in concave or thin features of the geometry

because the distance between two input edges is relative small.

In this paper, a posteriori method is employed to recover lost

boundaries by adding pseudo-points iteratively [7], [8].

1. Add pseudo-points

As shown in Fig. 2(c), after the boundary points are

generated according to the specified point spacing, the input

edges 5b and 11b have been converted into 5 17-P P , 17 6-P P

and 11 22-P P , 22 12-P P , respectively. And the lost edges in Fig.

2(d) are 17 6-P P , 11 22-P P , and 22 12-P P in practice.

Pseudo-points are produced at the midpoints of the lost

edges. For example, pseudo-point 69P is added at the

midpoint of edge 17 6-P P . Following by that, new Voronoi

diagram is established. Edge 17 6-P P is subdivided into

17 69-P P and 69 6-P P . For each new edge, further check

continues until there is no missing edge left.

Fig. 4(a) shows the triangular mesh after adding

pseudo-points, where black points 68P , 69P , and 70P

represent the added pseudo-points. Fig. 4(b) shows the

resulting mesh after eliminating the Voronoi vertices outside

the geometry.

2. Removal of pseudo-points

The addition of pseudo-points resolves the boundary loss

problem successfully, but also has a serious impact on the

accurate control of mesh density. Since the pseudo-point is

added at the midpoint of the lost edge, the accuracy of

geometric description for curved boundaries is decreased to

some extent. To resolve this problem, the added

pseudo-points should be removed again under the premise of

holding the integrity of geometry boundaries. In this paper,

several removal modes of pseudo-points are established by

defining different contracted points. Fig. 5 provides the

establishment of removal modes by taking the pseudo-point

69P as the example.

First of all, the Voronoi vertices whose dual Delaunay

triangles share the pseudo-point should be found out. The four

triangles sharing 69P are

69 35 17P P P , 69 18 35P P P ,

69 7 18P P P ,

and 69 6 7P P P as in Fig. 5(a).

(a) Add pseudo-points

(b) Delete outside vertices

(c) Remove pseudo-point

Fig. 4. Boundary integrity

ISSN: 2277-3754



139

(a) initial triangles (b) mode 1 (c) mode 2

(d) mode 3 (e) mode 4 (f) mode 5

Fig. 5. Pseudo-point removal

Table II. Point structures of each removal mode

Voronoi Mode 1 Mode 2 Mode 3 Mode 4 Mode 5

V1 P17, P6,

P7

P35, P17,

P6

P18, P35,

P17

P7, P18,

P35

P6, P7,

P18

V2 P17, P7,

P18

P35, P6,

P7

P18, P17,

P6

P7, P35,

P17

P6, P18,

P35

V3 P17, P18,

P35

P35, P7,

P18

P18, P6,

P7

P7, P17,

P6

P6, P35,

P17

Then, choose a contracted point from the forming points of

the Voronoi vertices connected to the pseudo-point. In

principle, every forming point of the triangles except the

pseudo-point to be removed can be considered as a contracted

point. Each of the original triangles provides an edge to form

a new triangle with the contracted point. In this case, five

removal modes can be established for the situation of Fig. 5(a)

respectively by taking point 17P , 35P , 18P , 7P , and 6P as

the contracting point, as shown in Fig. 5(b)-(f). The point

structures of the Voronoi vertices for these five removal

modes are listed in Table II. It can be seen that no matter

which mode is applied, the original four Voronoi vertices will

always be reduced to three ones. The next step is to select the

most reasonable removal way from the five modes in Fig.

5(b)-(f). This paper employs two judgment criteria to realize

the selection operation: Jacobian and minimum angle.

(1) The Jacobian

The Jacobian matrix [15] is the basic indicator to reflect

mesh quality. Fig. 6(a) shows a triangle with three forming

points denoted as 1 1 1, P x y , 2 2 2, P x y , and

3 3 3, P x y ordered in the counterclockwise direction. At

each point iP , the Jacobian matrix is expressed as:

Ai k i k

j i j i

x x y y

x x y y

(7)

det(A)Jacobian (8)

(a) a triangle (b) the empty polygon

Fig. 6. Insertion of interior point

where , , 1, 2, 3i j k . If 1i , then 2j and 3k .

If 2i , then 3j and 1k . If 3i , then 1j and

2k .

For a triangle in two-dimensional space, three Jacobian

values, one each for one of its three points, can be calculated

by “(7)” and “(8)”. Among these three Jacobians, compute

the minimal value as the Jacobian of the triangle. If the

Jacobian of a triangle is larger than zero, it signifies that its

orientation is positive and satisfies the requirements of mesh

generation. If the Jacobian is equal to zero, it signifies that the

triangle is degenerated into a single straight-line. If the

Jacobian is negative, the triangle is distorted or overlapped.

Therefore, only the triangles with 0Jacobian are

permitted for a finite element mesh. Compute the Jacobian

values of the triangles for the five modes in Fig. 5(b)-(f). The

result shows that 35 7 18P P P in mode 2 and 7 18 35P P P in mode

4 both have 0Jacobian , which exceeds the acceptable

range and is not permitted for a finite element mesh. By

Jacobian criterion, mode 2 and 4 are excluded. The follow-up

work is focused on selecting the best mode from modes 1, 3,

and 5.

(2) The minimum angle

The minimum angle is another important indicator for the

evaluation of triangular mesh quality. In this paper, the

minimum angle of each mode is defined as the minimal value

among the minimum angles of all the triangles in that mode.

min min( ) 1,2 ... i i n l l (9)

Compute the minimum angle of each possible mode by

“(9)”. The one which yields the largest value is selected as the

actual way. Computing result shows that the minimum angles

of modes 1 , 3 , and 5 are o16.31 ,

o16.31 , and o11.49 ,

respectively. So, exclude mode 5 and accept mode 1 in

sequence. Fig. 4(c) shows the resulting mesh after boundary

integrity, which can precisely capture the boundary features

of the analyzed geometry.

F. Generation of interior points

This paper uses two methods to produce interior points: the

direct method and pre-test method.

1. The direct method

The direct method generates the interior points according

to the specified point spacing of the points around them

ISSN: 2277-3754



140

directly. As shown in Fig. 6(a), an interior triangle has three

forming points labeled as 1P ,

2P , and 3P , its interior point

can be computed by four steps:

(1) Compute the centroid of the triangle, labeled as mP .

(2) Compute the practical point spacing at mP by

interpolation. 3

1

3

1

( )

'( ) 1,2,3m i i

im

i

i

P P A

ps P i

A

(10)

where is an adjust parameter ranging from 0.5 to 1.0 and

iA represents the area of the triangle formed by mP and

another two points, as shown in Fig. 6(a).

(3) Determine the minimal value of 1ps P , 2ps P ,

and 3ps P , denoted as minps . minps is

considered as the required point spacing at mP .

(4) Check whether ' mps P is larger than (min)ps . If

yes, accept mP . If no, reject mP . If mP is accepted, insert it

into the present Voronoi diagram. At the same time, a point

spacing value, ( )mps P , is assigned by interpolation.

3

1

3

1

( ( ))

( ) 1,2,3i i

im

i

i

A ps P

ps P i

A

(11)

2. Pre-test method

Pre-test method generates interior points through a

previous test process. For the triangle of Fig. 6(a), pre-test

method is performed by the following steps:

(1) Compute the centroid of the triangle, mP .

(2) Suppose that mP is required for generation. Insert it to

the present Voronoi diagram, delete the Voronoi vertices

whose circumcirles contain mP and construct an empty

polygon, see Fig. 6(b), where 'iP denotes the points on the

empty polygon.

First, compute the practical point spacing of mP by

1

1'( ) '

n

m m i

i

ps P P Pn

(12)

Then, pick out the minimum value of the specified point

spacing among the n points 'iP , denoted as minps .

Compare the values of '( )mps P and (min)ps . If

'( ) (min)mps P ps , accept mP to reduce the error.

Otherwise, reject mP . If mP is accepted, insert it into the

present Voronoi diagram and assign a point spacing as its

specified value during the subsequent procedures.

1

1( ) ( ')

n

m i

i

ps P ps Pn

(13)

3. The control on the number of cycles for inserting interior

points

When the interior points are inserted into a triangle lying on

the geometry boundaries, some undesired problems might be

considered. If a triangle has two points on the boundaries with

larger values of specified point spacing and one point in the

interior with a relatively smaller value of specified point

spacing, the practical point spacing at the centroid may be

increased undesirably due to the long length of boundary edge.

This will result in the ceaseless insertion of interior points and

thereby reducing the quality of boundary triangles, and even

leads to an endless loop of programming and the

non-convergence of solution. In order to avoid that problem, a

termination condition for the cycle control on inserting

interior points is employed, that is, the radius radio of the

circumcircle to the in circle of the triangle. In this paper, if a

triangle has the ratio larger than 2.0, it is considered that there

is no longer need to insert any interior points to it.

4. Smooth of interior points

Laplacian method [16] is the most commonly used method

for point smooth. It moves each point to the average position

of its neighboring points or elements in the most simplest and

straightforward way. Assume that the original coordinates of

an interior point iP are 0 0, iP x y . Its position coordinates

after smooth can be calculated by

0 0 0 0

1

1( , ) ( , ) ( ( , ) ( , ))

n

i i j i

j

P x y P x y C x y P x yn

(14)

where, ( , )jC x y and n respectively represent the center

and the number of the triangles connected to iP . is an

adjustment factor and set to 0.8 in this paper. Fig. 7(a) shows

the resulting mesh after inserting interior points using direct

method, consisting of 152 points and 227 triangles. Fig. 7(b)

shows that using pre-test method, with 229 points and 381

triangles. Both of these two meshes can realize the effective

control on the density distribution of the points and elements.

In comparison with the pre-test method, the direct method

occupies less storage space, spends less time and improves the

meshing efficiency significantly. Whereas the pre-test method

increases the programming work and meshing time due to the

operation of previous test. But, when pre-test method is used,

the density transition is rather smooth. Fig. 7(c) and (d)

provide the final meshes after smooth of the resulting meshes

in Fig. 7(a) and (b).

ISSN: 2277-3754



141

(a) Mesh with direct method (b) mesh with pre-test method

(c) Smooth of (a) (d) smooth of (b)

Fig. 7. Insertion and smooth of interior nodes

IV. CONTROL OF MESH DENSITY FOR

GEOMETRIES WITH SEVERAL SUB-DOMAINS

In practical engineering applications, it is usually required

to generate Delaunay triangular meshes for the geometries

with several sub-domains. Such meshes are required to meet

the following conditions in general: (1) Different

sub-domains have different density distribution; (2) The mesh

on the intersecting boundaries of two neighboring

sub-domains has to share the common points and edges; (3)

The points on the intersecting boundaries are conformal for

both of the local meshes of two associated sub-domains; (4)

The density transition of two adjacent sub-domains should as

smooth as possible. These conditions enhance the difficulty in

Delaunay triangulation for multi-domain geometry

considerably. In order to ensure the mesh conformity for the

geometries with several sub-domains, this paper established

the corresponding method and treatment way through

identification and treatment of overlapped edges. The edges

on the intersecting boundaries are treated as overlapped edges.

Special treatment procedures are carried out for the

conformity of the boundary points on overlapped edges.

A. Identification of overlapped boundary edges

When the two-dimensional geometric model to be meshed

contains several sub-domains, the boundaries for each

sub-domain should be input separately. For example, the

geometry as shown in Fig. 8(a) has three sub-domains. The

boundary edges of the first sub-domain are input as outer ring

1 2 3 4- - -P P P P and inner ring 8 7 6 5- - -P P P P . The second

sub-domain is input as outer ring 5 6 7 8- - -P P P P and inner ring

12 11 10 9- - -P P P P . The third sub-domain contains only one ring,

9 10 11 12- - -P P P P . First of all, the overlapped boundary edges,

i.e. the edges on the intersecting boundaries of two

neighboring sub-domains, have to be picked out, as the edges

on rings 5 6 7 8- - -P P P P and 9 10 11 12- - -P P P P .

(a) The input geometry

(b) Mesh after boundary integrity

(c) Mesh with direct method

(d) Mesh with pre-test method

Fig. 8. Treatment of overlapped boundaries for the

geometry with several sub-domains

ISSN: 2277-3754



142

B. Treatment of overlapped edges

It should be noticed that the points on the intersecting

boundaries are shared by both of the adjacent sub-domains

but the edges are overlapped and each belongs to its own

sub-domain only. For this reason, in the subsequent meshing

procedures, overlapped boundaries have to be treated

particularly.

(1) In the process of identification and insertion of

boundary points on the overlapped edges based on point

spacing, the points are added only once, but the edges are

updated repeatedly. For example, 15P is added on the pair of

overlapped boundary edges 6 5-P P and 5 6-P P . Once 15P is

inserted, the pair of overlapped edges 6 5-P P and 5 6-P P are

divided into two new pairs.

(2) During the process of boundary integrity, the insertion

and removal of pseudo-points are performed once but the

updating of edges is repeated.

Fig. 8(b) shows the resulting mesh after boundary integrity.

Each point on the overlapped edges is unique. But, the

overlapped edges are updated with the addition of new points

correspondingly. The newly generated edges on the

overlapped boundaries are also overlapped. Fig. 8(c) and (d)

provide final meshes when direct method and pre-test method

are used. In these two meshes, the points on the overlapped

boundaries are conformal for their two connected

sub-domains.

V. APPLICATIONS

The whole procedures and key techniques of Delaunay

triangulation illustrated above are implemented by

object-oriented C++ programming. In this section, three

examples are provided to demonstrate the reliability and

effectiveness of the system developed in this paper. The first

example is about a geometrical model containing curved

boundaries and two sub-domains, as shown in Fig. 9(a). The

thick lines indicate straight boundaries and thin lines indicate

curved boundaries. According to the curved and straight

status, the geometry boundaries are decomposed into five

segments, where 1S , 3S , and 5S are curved so that

B-splines should be constructed for them. Fig. 9(b) shows the

boundary contour map of the geometry after generating

boundary points. Fig. 9(c) shows the final mesh using direct

method, containing 1,151 points and 2,215 triangles. Fig. 9(d)

shows the final mesh using pre-test method, with 1,906 points

and 3,725 triangles. Fig. 10 presents the second triangle

meshing example, containing nine sub-domains and three

factors required to be refined particularly: point 1P , edge

2 3-P P , and segment 1S . Segment 1S is a curved line formed

by eight points, and required to construct a smooth curve

using B-splines, as shown in Fig. 10(b). In addition, there are

a lot of overlapped edges on the intersecting boundaries

between each pair of neighboring sub-domains. These

overlapped edges require treated specially. Fig. 10(c) shows

the final mesh with the interior points generated using direct

method, consisting of 2,754 points and 5,430 triangles in all.

Fig. 10(d) provides the locally amplified views of the three

refinement factors. The third example is to generate triangular

mesh for a complicated geometrical model containing a single

domain and several curved boundaries, as shown in Fig. 11(a).

Fig. 11(b) shows the final mesh, consisting of 5,746 points

and 10,908 triangles. Obviously, the triangular meshes

generated by the system developed in this paper are able to

capture the boundary features of the geometry precisely. The

density of mesh elements is distributed reasonably according

to the specified point spacing.

VI. CONCLUSION

This paper presented the implementation of Delaunay

triangulation to generate high quality meshes with prescribed

heterogeneous spacing. The applications proved that our

developed system was able to generate high-quality and

reasonable-density triangular meshes for input geometries.

(1) Our system generated boundary points by recursively

inserting midpoints according the specified point spacing.

B-splines were constructed for the curved boundaries. In this

way, the reasonable distribution and smooth transition of the

points on geometry boundaries were achieved. And the

boundary mesh was able to capture the geometric features

especially curved features of the analyzed model precisely.

(a) The input geometry (b) Boundary contour after

Generating boundary points

(c) Mesh with direct method (d) mesh with pre-test method

Fig. 9. Example 1

ISSN: 2277-3754



143

(a) The input geometry (b) construction of B-spline

curve

(c) Final mesh

(d) Amplified views of refinement factors

Fig. 10. Example 2

(a) The input geometry (c) final mesh

Fig. 11. Example 3

(2) Our system resolved boundary loss problem by adding

pseudo-points. Two criteria for selecting removal modes were

employed as well. By using this method, the lost boundaries

are recovered successfully. The mesh after boundary integrity

can capture the boundary features of the geometry accurately.

(3) Two methods were used to generate interior points:

direct method and pre-test method. Laplacian method was

used to smooth the interior points. The proper distribution and

conformal transition of interior points was realized

effectively.

(4) For the geometries with several sub-domains,

overlapped boundary edges were established. Corresponding

methods were established to treat the overlapped boundaries.

The reasonable distribution and excellent conformity of the

triangles and points between adjacent sub-domains are

implemented correctly.

VII. ACKNOWLEDGMENT

This research is supported, in part, by National Science

Council under contract Nos. NSC 101-2116-M-008-002 and

NSC 101-2811-M-008-067 with National Central University

and, in part, by National Central University under the 5/500

Fund.

REFERENCES [1] O. C. Zienkiewicz, “Achievements and some unsolved

problems of the finite element method,” International Journal

for Numerical Methods in Engineering, vol. 47, pp. 9-28, Jan.

2000.

[2] B. Delaunay, “Sur la sphere vide,” Bulletin of Academy of

Sciences of the USSR, vol. 7, pp. 793-800, 1934.

[3] C. L. Lawson, “Software for C1 surface interpolation,” in:

Mathematical Software III, J. R. Rice (ed)., Academic Press,

New York, pp. 161-194, 1977.

[4] D. Watson, “Computing the n-dimensional Delaunay

tessellation with applications to Voronoi polytopes,” The

Computer Journal, vol. 24, pp. 167-172, 1981.

[5] R. Sibson, “Locally equiangular triangulations,” The

Computer Journal, vol. 21, pp. 243-245, 1978.

[6] A. Bowyer, “Computing Dirichlet tessellations,” The

Computer Journal, vol. 24, pp. 162-166, Jan. 1981.

[7] N. P. Weather ill, “Delaunay triangulation in Computational

Fluid Dynamics,” Computers & Mathematics with

Applications, vol. 24, pp. 129-150, Sep. 1992.

[8] N. P. Weather ill, “The integrity of geometrical boundaries in

the two-dimensional Delaunay triangulation,”

Communications in Applied Numerical Methods, vol. 6, pp.

101-109, Feb. 1991.

[9] J. R. Shewchuk, “Delaunay refinement algorithms for

triangular mesh generation,” Computational Geometry-Theory

and Applications, vol. 22, pp. 21-74, May 2002.

[10] P. L. George, F. Hecht, and E. Saltel, “Automatic mesh

generator with specified boundary,” Computer Methods in

Applied Mechanics and Engineering, vol. 92, pp. 269-288,

Nov. 1991.

[11] P. L. George and H. Borouchaki, “Delaunay Triangulation and

Meshing: Application to Finite Elements,” Paris, Editions

HERM ES, 1998.

[12] I. J. Schoenberg, “Contributions to the problem of

approximation of equidistant data by analytic functions, Part A:

On the problem of smoothing of graduation, a first class of

analytic approximation,” Quarterly of Applied Mathematics,

vol. 4, pp. 45-99, 1946.

[13] C. De Boor, “On calculating with B-splines,” Journal of

Approximation theory, vol. 6, pp. 50-62, 1972.

[14] M. G. Cox, “The numerical evaluation of B-splines,” IMA

Journal of Applied Mathematics, vol. 10, pp. 134-149, 1972.

http://www.citeulike.org/user/wparker/author/Schoenberg:IJ

ISSN: 2277-3754



144

[15] P. M. Knupp, “A method for hexahedral mesh shape

optimization,” International Journal for Numerical Methods in

Engineering, vol. 58, pp. 319-332, Sep. 2003.

[16] D. A. Field, “Laplacian smoothing and Delaunay

triangulations,” Communications in Applied Numerical

Methods, vol. 4, pp. 709-712, 1988.

AUTHOR’S PROFILE Correspondent author: Gour-Tsyh Yeh

Education:

1967 – 1969: Ph.D. Hydrology, Cornell University

1966 – 1967: M.S. Hydraulics, Syracuse University

1960 – 1964: B.S. Civil Engineering, National Taiwan University

Research area:

Hydrology, environmental fluid mechanics, hydraulics, and water

resources, physics-based first principle approaches of watershed

modeling, coupled surface and subsurface flow and transport processes,

geochemical kinetics, biodegradation and micro-organism/geochemical

interactions, geochemical equilibrium modeling, multi-phase flow and

transport in both fractured and porous media, development of

innovative numerical algorithms, and computational fluid dynamics.

Experience:

2010-Date NSC Chair Professor, Graduate Institute of Applied Geology,

National Central University, Jhongli, Taoyuan, Taiwan

2000-2010 Provost Professor, Department of Civil and Environmental

Engineering, University of Central Florida, Orlando, FL.

1989-2000 Professor, Department of Civil and Environ. Eng.,

Pennsylvania State University, University Park, PA

1977-1989 Senior Research Staff (1983-1989); Research Staff II

(1977-1983) Oak Ridge National Laboratory, Oak Ridge, TN

1973-1977 Senior Hydraulic Engineer (1973-1975); Senior

Environmental Engineer (1976-1977), Stone and Webster

Engineering Corporation, Boston, Massachusetts

1975-1976 Senior Engineer Tetra Tech, Inc., Pasadena, California.

1972-1973 Senior Engineer, Ebasco Services, Inc., New York, New York.

1971-1972 Visiting Scientist on NRC Research Associateship, NASA,

Houston, Texas

1971-1971 Visiting Associate Professor, Dept. of Civil Engineering,

National Taiwan University, Taipei, Taiwan

1969-1971 Research Associate, Dept. of Civil Engineering, Cornell

University, Ithaca, New York

1967-1969 Research Assistant Dept. of Civil Engineering, Cornell

University, Ithaca, New York

1966-1967 Research Assistant, Dept. of Civil Engineering, Syracuse

University, Syracuse, New York

Achievements:

(1) Published two books and more than 400 papers, technical reports, and

conference proceedings; highest citation=435; Number of

publications with over 100 citation = 7; H-Index = 26.

(2) Developed over 100 physics-based process-level computational

models: many of which has been adopted as the de facto standard by

academic communities, federal agencies, and industries

(3) Conducted more than 40 interdisciplinary research projects with a

total funding of over $8,000,000

(4) Received Presidential PIP Award, Martin Marietta Publication Award,

NRC Research Associateship, NRC Senior Research Associateship,

PSES Outstanding Research Award, UCF Distinguished Researcher,

UCF Graduate Teaching Award, NTU Outstanding Alumni

(5) Consultant, IAEA, Unite Nations

(6) Completed over 40 consulting jobs for over 30 clients

(7) Offered short courses, seminars, and workshops many times on

subsurface flow and reactive chemical transport

(8) Organized and chaired more than 10 salient national and international

symposia

(9) Advised and supervised over 25 Ph.D. and M.S. students at University

of Central Florida and Penn State in 21 years

First author: Lu Sun

2012-Date Post-doctoral researcher, Graduate Institute of Applied

Geology, National Central University, Taiwan

2008-2012 Ph.D. Materials Science and Engineering, Shandong

University, China

2006-2008 M.S. Materials Science and Engineering, Shandong

University, China

2002-2006 B.S. Materials Science and Engineering, Shandong University,

China

Research area:

Finite element mesh generation

Third author: Fang-Pang Lin

Education:

1991-1995 Ph.D. Department of Civil Engineering, University of

Swansea Wales

1990 -1991 M.S. Department of Civil Engineering, University of

Swansea, Wales

1984 -1987 B.S. Department of Civil Engineering, National

Chiao-Tung University

Research area:

Computational Fluid Dynamics, Finite Element Analysis, Multigrid

Methods, Parallel and Distributed Computing, Optimization Design,

and Grid Computing

Working Experiences:

2003-Date Research Scientist, Division Manager, Grid Computing

Division, NCHC, Taiwan

2000- 2002 Associate Research Scientist, Software Development Division,

NCHC, Taiwan

1996-1997 Visiting Research Scientist, Oxford University Computing

Laboratory, UK

1995-1996 Postdoctoral research, Department of Civil Engineering,

University of Swansea Wales, UK

Achievements:

(1) Published published more than 40 papers and reports.

(2) Received the highest technology development award by Executive

Yuan, Republic of China (Taiwan)

(3) Organized and chaired more than 20 international conferences

(4) Deputy Chair, PRAGMA (Pacific Rim Applications and Grid

Middleware Assembly) Steering Committee; Steering Committee

Member, RCN (Research Coordination Network); Steering Committee

Member, GLEON (Global Lake Environmental Observatory Network);

Member of Advisory/Expert Panel of MIMOS (Malaysia Institute of

Microsystems); Steering Committee Member, ISC (International

Supercomputing Conference)

Fourth author: Guoqun zhao

Education:

1988-1991 Ph.D: Department of Materials Engineering, Shanghai

Jiaotong University, China

1983-1986 M.S. Department of Materials, Shandong University, China

1979-1983 B.S. Department of the second Mechanical Engineering,

Shandong University, China

Research area:

Plastic forming theory and technology, mold design and manufacture,

numerical simulation of plastic forming, mold CAD/CAM/CAE, mold

design and optimization, product/mold rapid design and manufacturing

process

Working Experiences:

2006-Date Dean, professor, doctoral tutor, School of Materials Science

and Engineering, Shandong University, China

2002-2005 Director, Department of Science and Technology, Shandong

University

2001-2001 Senior visiting scholar, Department of Mechanics and

Materials, Wright State University

1997-2001 Professor, doctoral tutor, director of mold center, Department

of Materials Science and Engineering, Shandong University

1993-1997 Postdoctoral researcher, Department of Mechanics and

Materials, Wright State University

1991-1993 Associate professor, Department of Materials Science,

Shandong University

Achievements:

(1) Published 4 classics and manuals, and published more than 200 papers.

(2) Obtained more than 10 patents, register 3 kinds of software.

http://www3.interscience.wiley.com/journal/117919660/home



ISSN: 2277-3754



145

(3) Hosted more than 20 projects, containing the National Science and

Technology Support Program, the National Outstanding Youth

Foundation, the National Natural Science Foundation of China, the

provincial (ministry) level research projects and enterprises

commissioned project etc.

(4) Obtained the second prize of National Science and Technology

Progress award in China. Obtained 18 types of provincial and

ministerial level scientific and technological achievement awards.

volume 3, issue 3, september 2013 an automatic delaunay ... 3/issue 3/ijeit1412201309_22.pdf ·...

Documents